[Mitgcm-support] Re: Releases, branches, etc...

[mitgcm-support at dev.mitgcm.org]

We met on January 30th in the Charney library to discuss and
agree on the policy for the release1 branch.

1. After three of us badgered Chris, we agreed that the original
branch "release1-branch" had been made at the wrong point and
was undocumented. I created a branch "release1" from c43 that
was then updated to exactly replicate Chris's branch.

2. We agreed to bracket changes on the main trunk and to merge
selected brackets into the release1 branch. This provides
continuity between releases on a single branch and allows
us to document changes between each release.

3. Chris adamantly defined release1 as be the state of the
code given by -D "12:45 11 Jan 2002" along with tutorials and
then yet to be made fixes for GM. We persuaded him that there
were other bug fixes that should be added but we conceded to
not include new code. 

Chris suggests it would take a month to make the merge. It in fact
takes five commands, one per bracket, to make the merge. I wrote
out the CVS instructions twice in the the thread "Release1
branch". It seems to me, however, that there is a reluctance to
follow the policy rather than an inability to comprehend/remember
the decisions of our January 30th meeting?

If we do agree on just one continuous branch for release1 then
the bracketed changes can be merged in very easily using cvs update.

However, I also have a clear recollection that Chris was adamant
that the release1 would be based on checkpoint43a plus GM fixes only.
i.e.   release1-beta1 + diffs bracketed by c44d_pre/post.
Q) Is this no longer the case?

Now it appears that the more complete code is to be made
the release. May I point out that most of the diffs listed in Chris's
fig. are not bug fixes. However, the advantage of this approach is
that it requires only one CVS merge of the diffs between
c43a and c44d_post which will make the release1 branch identical
to the c44d_post tag and will include the tutorials. This is certainly
the easy option but I seem to have missed why "release 1" is now
all encompassing?

I personally prefer for everything upto c44d_post being merged
into the branch. It avoids this rather arbitrary definition of R1 based
on a time of day.

I do not want a second branch for r1b2. Note that under cvsweb
there is no documentation of what is different between r1b1 and r1b2:
http://mitgcm.org/cgi-bin/cvsweb.cgi/MITgcm/model/src/?sortby=date&only_with_tag=release1_final_v1
while merging forces us to document what has happened:
http://mitgcm.org/cgi-bin/cvsweb.cgi/MITgcm/model/src/?sortby=date&only_with_tag=release1

Also, this is not the "final" release of release1 and so let's not start
using such words.

A.



A.